Content area
Agent-based modeling (ABM) plays a critical role in complex systems research by allowing researchers to examine how individual-to-individual interactions collectively give rise to group-level and system-level behavior. However, fields ranging from socio-environmental systems to tumor biology to traffic modeling have increasingly sought to model interactions between systems of different scales. Multi-level agent-based modeling (ML-ABM) extends classic ABM techniques to meet this need. Despite this growing demand, multi-level modeling techniques introduce significant complexity into the modeling process and have yet to see widespread adoption among ABM practitioners. We introduced the LevelSpace extension for the widely used NetLogo ABM platform to make ML-ABM easily accessible to ABM researchers by leveraging NetLogo’s core “low floor, high ceiling” approach.
This dissertation builds on that work, showing how researchers can model multi-level phenomena by creating nested hierarchies of agent-based models. It accomplishes this through a series of novel and illustrative case studies. Each begins with a classic, single-level agent-based model, and then extends it to a multi-level modeling system. In each case, we explore the different kinds of relationships that can connect models, with a particular focus on the amount of coupling and re-usability of the component models involved. We perform in-depth experiments and analyses of each model, demonstrating the techniques involved in analyzing ML-ABMs, comparing the behavior of ML-ABMs with single-level ABMs, and gaining new insights into the simulated systems. Finally, we demonstrate that ML-ABM offers powerful techniques for defining agent cognition in particular by allowing agents to model their environment and make decisions using subordinate ABMs. The case studies presented in this dissertation offer thorough yet accessible templates by which to guide other researchers in the design and analysis of multi-level agent-based models.
